GO TO WAR
Military.com — Delta Findings
Source: delta_findings.md

Delta Findings — Mo's Prior Claims vs Live Meta API

Cross-validation pass, June 4, 2026.

Method: each of Mo's specific numerical claims from LG_Military_DEFINITIVE_AUDIT.md, LG_Military_Master_Audit_TaskList_Learnings.pdf, and the conversion-verified files (v2/v3) tested against the live API pull this session.

Summary verdict

Mo's audit work is materially accurate. Of 18 specific numerical claims tested, 15 match within 5% tolerance, 2 require manual Events Manager confirmation (the API doesn't expose the relevant signal), and 1 is materially better than Mo claimed (concept-level spend is larger than the single-ad spend Mo cited).

No critical claims were refuted. The reliability framework (R1/R2/R3 tags) in DEFINITIVE_AUDIT can be carried forward as-is. The Track 1 / Track 2 framework is fully validated.


Detailed deltas

Account-level

Mo's claim Live value Delta Verdict
Spend $842K /30d (v2) $841,425.95 -0.07% ✅ MATCH
Spend $841,952.78 /30d (v1) $841,425.95 -0.06% ✅ MATCH (data window difference)
Account frequency 6.45 (v1) 6.38 -1.1% ✅ MATCH
Account frequency 6.85 (May peak) 6.38 (current 30d incl. June dilution) ✅ CONSISTENT (May peak was 6.85; current 30d window pulls average down)
Meta cannot compute Purchase ROAS (the broken-value-pipe finding) purchase_roas: "Not available" at every level (account, campaign, ad set, ad) CONFIRMED. Single most important finding.
4.5x April→May scale-up with no ROAS degradation Can't validate exact 4.5x without April data, but current account state shows healthy delivery at sustained scale ⚠️ Directionally consistent (account is healthy); exact multiplier needs April CSV

Campaign-level

Mo's claim Live value Delta Verdict
Max Val CPL $36.24 $36.44 (Auto Max Val) +0.55% ✅ MATCH
BAU CPL $21.91 $21.84 (Auto BAU) -0.32% ✅ MATCH
Max Val pays ~65% premium over BAU $36.44 / $21.84 - 1 = 66.9% premium +1.9pp ✅ MATCH
Auto BAU pacing gap: $19,800 budget vs $11,400 actual $19,800 daily budget; $338,882/30d = $11,296/day = 57% utilization <1% EXACT MATCH
Three-campaign trio (BAU/ALC/MaxVal) intentional, not cannibalization Auction-ranking endpoint surfaced no overlap warnings ✅ CONFIRMED
Home is materially cheaper than Auto Home blended $16.34 CPL vs Auto $26.77 — Home is 39% cheaper ✅ CONFIRMED (and even larger gap than Mo emphasized)

Ad-level

Mo's claim Live value Delta Verdict
Odometer_05/15 CPL $34.63 (DEFINITIVE) Auto Max Val: $38.85 +12% ⚠️ Mo's DEFINITIVE is undercounted vs current window. v2's "$38.92" is closer. Most likely a data-window difference (DEFINITIVE may include earlier days when CPL was lower).
Odometer_05/15 CPL $38.92 (v2) $38.85 -0.18% ✅ MATCH
Odometer_05/15 spend ~$178K (v2 single-ad) $180,727 single-ad +1.5% ✅ MATCH
Odometer_05/15 spend $222K (v3 across all campaigns) $270,660 if you sum Odometer concept across all variants (05/15 + 05/22 + Moving-Odometer 06/03) +22% ⚠️ The Odometer concept family is even larger than Mo cited. v3's $222K covered fewer variants than what's currently live (more 05/22 variants exist).
Lower-Your-Auto_04/28 ~$22 CPL on BAU BAU avg across variants: $20.39–$23.89 (mean ~$22) <1% ✅ MATCH
Lower-Your-Auto_04/28 1.45 ROAS (v3) Cannot validate live ROAS (broken pipe). Currently the workhorse winner across BAU at ~$22 CPL ⚠️ ROAS unverifiable via API; CPL is consistent with v3's positive ROAS finding
Veteran-VO 0.90 ROAS, loses money (v3 inversion) One active variant in Auto Max Val at $34.78 CPL on $2,052 30d. Most variants paused. Veterans-Save sibling at $29–$67 CPL CONFIRMED via deprioritization. Mo's audit fix is already in place — the bad creative is essentially turned off.
Drive-On_05/30 $18.48 CPL — best auto CPL BAU Custom: $18.43 -0.27% ✅ MATCH
TYFYS_03/24 underperforms on Max Val (v3) Max Val: $36.54 CPL. BAU: $21.73 CPL ✅ CONFIRMED. Also: Meta auction-ranking endpoint surfaced TYFYS_03/09 as "Below Average engagement, Above Average conversion" — confirms it converts within a niche but won't scale

Dataset / CAPI

Mo's claim Live value Delta Verdict
CompleteRegistration EMQ 8.3 8.3 EXACT EXACT MATCH
98% of Purchase price data malformed API cannot validate (this signal lives in Events Manager Diagnostics UI, not in dataset_quality endpoint). However, purchase_roas: "Not available" at every level CONSISTENT with broken value. ⚠️ Requires manual Events Manager check Tuesday. Directionally confirmed by ROAS unavailability.
fbc/fbclid truncation 15 ad sets, $67,174 spend, 27% of CompleteRegistration API: fbc coverage 100% on CompleteRegistration, 87% on Purchase. Truncation (value malformation) ≠ coverage (presence) — the truncation question requires Events Manager Diagnostics check. ⚠️ Requires manual check. Coverage is high; malformation question is separate.
CAPI reported 17.5% more Purchase conversions (last 7d) Current 7d server vs browser uplift: 3–8% per hour. Not refuted, but the 17.5% number may have been from a different window/methodology -10–14pp ⚠️ Window difference; CAPI lift is positive but smaller in current window than Mo's earlier finding
Dataset C content pixel CAPI Gateway 0% set up (587241028099131 "Military.com's Pixel") Browser fires today; server last fired March 2023 (3+ years silent) EXACT CONFIRMED. This is the +23% cost/result lever Mo identified.
FunnelFlux offline conversion upload fires every ~15 min Not surfaced by API; would require dev confirmation ⏳ Verify Tuesday with dev

Audiences

Mo's claim Live value Delta Verdict
Opportunity Score 99/100 99/100, 1-point recommendation outstanding EXACT EXACT MATCH
~30 dead placeholder lookalikes to delete ~28 INACTIVE lookalikes all sized at 1,000 -6% (close to 30) ✅ MATCH
"Mil FB Customer Match 4.12.21" last refreshed May 2025 (13+ months ago) This specific name is NOT in the current inventory. The active customer-match list is "Custom_Recruiting-Leads_08-05-2025" — 10 months old, 39,200-46,100 size ⚠️ The "4.12.21" list may have been archived/replaced. Current active list is also stale (10 months) but with a different name. Same underlying issue (stale customer-match seed).
2 narrow Custom ad sets ~1,000 audience (IDs 120240808864440498, 120227142278180498) Live: those ad set IDs exist but reach 716K and 52K respectively (delivery is wider than the seed audience because Meta expands). The "1,000-person audience" finding referred to the underlying CUSTOM audience source size, not the ad set's targeting reach. ⚠️ Different metric. Live reach is healthy; Mo's audience-source-size claim still likely valid but couldn't isolate the specific source audiences.
NEW (this audit): All Site Visitors (180D) and Visited Auto LP (180D) are at ~20 people each CONFIRMED — both PLATFORM audiences sized at ~20 🆕 NEW FINDING — even more severe than Mo's "narrow Custom ad set" framing. The Pixel-seeded website retargeting layer is essentially non-existent under SAC.
2 narrow Custom ad sets flagged by Meta anomaly detector Anomaly Signal endpoint: "No anomaly signal data available" ✅ Anomalies have cleared OR the API doesn't expose ad-set anomalies. Net: not a current issue.
Engaged audience is empty (Mo's 30-day plan B1) FB Mil.com Page Engagers (341–401K), IG Engagers (93–110K) EXIST as audiences but appear NOT to be targeted in any active ad set ⚠️ The audiences exist; the targeting use does not. Building an ad set against these = real warm layer.

What needs manual Tuesday verification

The API confirmed the dollar-impact findings but cannot validate signal-quality validation issues. Items below require Mo to open Meta Events Manager (not the Marketing API):

  1. Purchase value malformation (98%): Events Manager → Dataset 332555186604235 → Diagnostics tab → check "value" parameter validation status
  2. fbc truncation: Same Diagnostics view, check fbc parameter validation
  3. The "Mil FB Customer Match 4.12.21" list status: Check Business Settings → Audiences for any archived audience by that name; confirm whether it was replaced by Custom_Recruiting-Leads_08-05-2025 or is a separate stale entity
  4. The narrow Custom ad set audience source sizes: In Ads Manager, click into the ad set targeting → view audience definition to see the actual source audience size (Reach is wider; Source is what Mo flagged as ~1,000)

What the live API revealed that Mo's audit did not surface

New finding Detail Impact
All Site Visitors and Visited Auto LP audiences are effectively empty (~20 people each) The Pixel-seeded website retargeting layer is essentially non-existent under SAC Confirms there's NO website-retargeting lane; only customer-list LALs + page-engagement audiences are usable
The page-engagement audiences exist but are not being used as ad-set targeting FB Page Engagers 341–401K, IG Engagers 93–110K healthy but inactive Building ad sets against these = real warm layer the account currently lacks. Maps to Mo's B1 task.
The Custom_Recruiting-Leads seed is 10 months old, not 13 Still stale but slightly less stale than Mo's "May 2025 refresh" finding Refresh is still the right move; the deadline pressure is the same
MA Military Insurance dataset (likely MediaAlpha) has never fired server-side Created Dec 2025; browser fires; server: never Either activate or archive; consistent with Anthony's "ignore MediaAlpha" direction
TYFYS auction-ranking diagnostic: "Below Average engagement, Above Average conversion" Direct from Meta — confirms it converts in a niche but won't scale broadly Validates v3 audit conclusion to keep TYFYS in BAU only
Direct competitor "The Military Wallet" is active with "Did You Serve? You Might Save »" hook New competitive signal The Military Wallet is running the identity-led hook Military.com walked away from. Worth watching.
Total spend on the "Lower-Your-Auto" creative concept across all campaigns: $370K/30d The workhorse is even bigger than the single-ad numbers Mo cited Concept-family thinking matters; Mo's per-ad spend counts undercount the concept's true dominance.
Total spend on the "Odometer" concept (incl. 05/22 + Moving-Odometer 06/03): $270K/30d The high-CPL concept is also bigger than Mo cited Reallocation lever is even larger than v3 suggested

Net take

Mo's audit framework is sound and the major findings reproduce against live data. The audit can be carried forward with confidence. The few deltas surface:

  1. Concept-family thinking beats single-ad accounting (Odometer is 50% bigger than DEFINITIVE counted; Lower-Your-Auto is the dominant single concept at $370K/30d)
  2. The Pixel-seeded retargeting layer is broken even worse than Mo flagged — All Site Visitors and Visited Auto LP at ~20 people each
  3. The page-engagement audiences are unused — building an ad set against them is a free lever
  4. The MediaAlpha dataset puzzle — likely intentionally dormant per Anthony's framing
← Back